The 90-Day Return Window: Why Velocity Beats Volume in Abuse Detection

Watch the rate, not the count. A customer who returns 3 items in 90 days is a different signal from one who returns 3 items in a week, and a customer who returns 20 percent of everything is different from one who returns 60 percent. The 90-day rolling window is the standard because it is long enough to see a pattern and short enough to act on one. Track returns per customer, return rate, and time between order and return, and flag accounts that cross your threshold on two of the three. Volume alone will flag your best customers. Velocity will not.

Volume flags your best customers

The most common mistake in return abuse detection is counting returns. A customer who returns 12 items in a year looks like a problem, until you notice they bought 60. A 20 percent return rate is normal in apparel. The customer who returns 3 items in a year looks harmless, until you notice they bought 4. Counting returns without counting purchases flags the customers who spend the most, and misses the ones who return the most.

Velocity fixes the first half of that mistake. It measures the rate, not the count: how many returns, over how long, relative to how much was bought. A customer who returns 3 items in a week has higher velocity than one who returns 3 items in 90 days, even though the count is the same. A customer who returns 60 percent of everything, month after month, is a different signal from one who returns 20 percent once. The rate is the pattern. The count is just a number.

The 90-day rolling window is the standard because it matches how the abuse actually happens. Wardrobing, buy-wear-return, clusters around events: a wedding, a holiday, a weekend. A 30-day window is too noisy, one holiday order can look like abuse. A 12-month window is too slow, the abuse is over before the flag fires. Ninety days is long enough to see the pattern and short enough to act on it, and it is the window most return scoring systems use.

What to measure, and what to ignore

Three measures carry the signal. Return rate per customer, the percentage of what they buy that comes back, over the rolling 90 days. Return count in the same window, which catches the customer who returns a lot of cheap items. And time between order and return, which catches wardrobing: the item that comes back 5 days after the order, after a weekend, with signs of wear.

Two measures look useful and mostly are not. Total lifetime returns, which punishes loyalty. And return reason codes on their own, which are self-reported and often wrong. A customer who returns everything as defective is not necessarily lying about the defect, and a customer who returns everything as wrong size is not necessarily telling the truth. The reason code is a hint, not evidence.

The flag should trigger a review, not a ban. Bracketing, buying multiple sizes of the same item and returning all but one, looks fast on velocity and is legitimate shopping. A customer who buys for a family, or a stylist who buys for clients, will have a high return rate for a good reason. The score is the start of a decision, not the decision. The decision is a human looking at the pattern and asking whether it is abuse or just a customer who shops differently.

The threshold, and the response

The threshold is a policy decision, not a formula, and it should be reviewed quarterly. A reasonable starting point is two of three: a return rate above your store average by a wide margin, more than a set number of returns in 90 days, and a short time between order and return. Two of three keeps the false positives down, because a single measure can always be explained away.

The response is graduated, and the first step is a warning, not a ban. A message that explains the return rate, the cost, and the policy, in plain language, stops most misuse on its own, because the customer did not know anyone was counting. Restrict next: no free return shipping, a shorter window, store credit instead of a refund. Ban only if the behavior continues after the restriction. The NRF puts fraudulent and abusive returns at about $103 billion a year, and wardrobing is reported by 60 percent of retail executives. The money is real, and the warning is the cheapest way to recover it.

Frequently asked questions

What is return velocity?

The rate of returns over time, usually measured in a 90-day rolling window: how many items a customer returns, and how quickly, relative to what they buy. A customer who returns 3 items in a week has higher velocity than one who returns 3 items in 90 days, even though the count is the same.

Why 90 days?

It is long enough to see a pattern and short enough to act on one. A 30-day window is noisy, one holiday order can look like abuse. A 12-month window is too slow, the abuse is over before the flag fires. Ninety days is the window most return scoring systems use, and it matches how wardrobing actually happens, around events and seasons.

What threshold should I use?

There is no universal number. Start with two of three: a return rate above your store average by a wide margin, more than a set number of returns in 90 days, and a short time between order and return. The threshold is a policy decision, not a formula, and it should be reviewed quarterly.

Does velocity flag good customers by mistake?

Volume does. Velocity usually does not, because good customers return slowly and spread out. The false positives come from bracketing, buying multiple sizes and returning all but one, which looks fast but is legitimate. That is why the flag should trigger a review, not a ban.